Inspec Insights

Interpreting ANOVA GR&R Results

By Ed Pietila, Quality Manager 

Graphic Results


Interpreting Graphic Results

Header Information:

Part Number: Obvious

Gage Name:  Should be the Gage Name and Gage ID for the equipment used to measure this feature (eg. CMM I-005)

Date of Study: Study Completion Date

Reported by: Person who is creating the ANOVA report

Tolerance: Tolerance Range for the specified measurement

Misc: Enter the Nominal Value of the Measurement



 Components of Variation:

This Chart show the various Components of Variation, based on % Contribution (blue), % Study Variation (Red) and % Tolerance (Yellow).  The different Components are:

  1. The Overall Gage R&R
    • Combination of Repeat. and Reprod. Through complex calculation
    • Needs to be below 20% to be acceptable.  The Yellow Graph is normally the most important
  2. The Repeatability
    • This is showing how the Program/Fixture repeats
    • This should be below 20% for any chance of the Gage R&R to be below 20%
  3. The Reproducibility
    • This is showing how the Operators compare for the specific parts
    • This should be below 20% for any chance of the Gage R&R to be below 20%
    • In this example above, the operators were very comparable in their results
  4. Part-to-Part Variation
    1. This is showing how the different part values compare as a percentage of Contribution, Study Variation, and Tolerance
    2. Ideally, this should be the highest graph for a very good GR&R

In a study with good results, the Gage R&R yellow value should be very low and the Part-to-Part be the highest contributor.

Data by Part

This Chart shows:

  1. The Average value for each of the 10 parts (based on the 9 measurements of it – 3 from each operator) which is the line.
  2. Each grey dot indicates a reading.
  3. For a 10-3-3 GR&R (10 Parts, 3 Operators, 3 Tries) there should be 9 grey dots per part.  Some may have the same values
  4. For an ideal GR&R, all the 9 grey dots for each part should be on top of each other

The R Chart (Range) show:

  1. The Range of values for the  10 parts (shown at bottom)  for each Operator (shown at top)
    • Left side shows the levels – in this case 0.000, 0.001, and 0.002
    • The blue dots show the range for each part number
    • Right side shows:
      • The Upper Control Level (UCL) to show that the process is under control
      • The Average Range for all parts
      • The Lower Control Level (LCL)
  2. The graphs for each operator should be as low as possible
    • The highest graph (by operator) shows which operator needs to improve their repeatability of part placement

Data by Operator

The Data by Operator (“Box and Whiskers Graph”) shows:

  1. The circle with the cross in it is the average (Mean) for all parts for each operator
  2. The horizontal line in the box is where 50% of the results are above the line and 50% below (this is different than the Mean, as indicated above, it is the Median)
  3. The blue box contains the mid 50% of all readings
  4. The vertical line (whisker) above the box indicates the top 25% of all readings
  5. The vertical line (whisker) below the box indicates the bottom 25% of all readings

The smaller the lines and boxes, the tighter the range of values (based on scale on left side of graph)

X-Bar Chart by Operator

The Xbar Chart by Operator shows:

  1. Xbar (Average) by operator for each part in the 10 piece study
  2. The left side shows the scale for the average readings
  3. The right side shows:
    • The Average of the Averages (Xbar-bar) for all parts, all operators
    • The UCL based on Standard Deviations – indicating where a process becomes “Out of Control”
    • The LCL, also based on Standard Deviations and also indicating “Out of Control” limit

Ideally, all 3 graphs (1 for each operator) should be identical.  As long as the 3 operators have identical (or close to identical) graphs, points being outside the UCL or LCL only indicates a problem with the part creation process and does not affect the GR&R but only indicates the creator of the parts needs to improve their process.

Part * Operator Interaction

The Part * Operator Interaction Graph shows:

  1. The average values (by operator) for each part, similar to the Xbar chart above, but with the 3 operator graphs overlaid (all on top of each other)
  2. The left shows the scale for the readings
  3. The right side is the “key” to indicate which operator is which graph

Ideally, all 3 graphs will lie on top of each other.  Any operator point deviating from the other points (example above – part #9 shows operator 2 & 3 very close to the same reading of about 0.1174 and operator 1 having a reading of about 0.1158) indicates there is a problem with operator #1 placement of the part in the fixture (if on a CMM or Vision System) or their measurement method (if with hand gages, like micrometers).

Text Results

Of all the Text Results indicated, at our level of comprehension we only need to focus on the first and last sections.

The first section indicates:

  1. The Gage Name – we would put “CMM I-005” or whatever the equipment and GageID is
  2. Date of study completion
  3. Who created the study
  4. The Tolerance range used for the study
  5. The nominal value for the feature

The last section begins where it indicates “Gage R&R”.

The first part of that section is showing the Variation Compensation (from ANOVA – Analysis of Variation) and doesn’t concern our results at this level

The next line shows our Process tolerance (the range of tolerance allowed for this feature measurement)

The last section shows the different contributions to the final value. 

In this example, the Gage R&R value is 92.12% (Bad – should be under 20%) as seen in the right column under %Tolerance. This value is the combination of the Repeatability, Reproducibility, and Operator contributions.  This example shows that all of the variation is from the Repeatability.  Although the Part-to-Part is shown in this table, it is not used in the calculation of final GR&R for the reason indicated when describing the graphs above.

The Distinct Categories comes more into play if the parts are coming from different molds and there is an obvious split in the data from the various parts.  For our process, we are not concerned with that information.